Build regression model from formation signals

Create a simple linear regression model that maps formation signals to end-of-life features.

Created: 4/18/2021

Authors: Valentin Sulzer, Andrew Weng

Point to the dataset

Load the correlation data

Define the models

Single feature

Set up and fit/predict/score

Take a look at the results list variable.

It is a list of dictionaries. Each dictionary holds the output for a single cycle index.

Re-package the data to become easily plottable

Repackage the data to index by feature, which makes it easier for plotting.

There is probably a more "pandas-y" way to do this.

Visualize the results

Multiple features from formation

Data analysis

Select relevant formation features

Look for correlations

Select and plot features with abs(corr) > 0.6

with log

Training a regularized model

Individual features

with log

Manual hyperparameter loops

Manual, high corr features

Manual, very high corr

Printing lots of models

PCA

Understanding the model

See https://scikit-learn.org/stable/auto_examples/inspection/plot_linear_model_coefficient_interpretation.html

Including features up to cycle 3

PCA

So now we have two interesting directions

Features for the paper